Change tempo - worse when high quality

Hi, I would like to change the tempo of several hours of recordings I made for a project management course, because I was over compensating on the virtual experience by talking too slowly, in some sections far too slowly.

I need to sync with video, so used the “high quality stretching” option to ensure an exact time compression (as per the documentation). However, interestingly, it caused much more distortion compared to not using it, at least on this kind of voice recording.

The following clip has some original audio, then with tempo sped up by 10%, then the same speed up of 10% but with “high quality stretching”. The last is significantly distorted.

From my experiments, the maximum increase in tempo of my audio I could get without noticeable changes was about 10%, always without using the high quality option. This might be a general rule of thumb for voice recordings.

On the other hand, if anyone knows of any other ways to increase the tempo of vocal recording more than 10%, or reduce the distortion when you do, any pointers would be greatly appreciated.

The “high quality” setting works best for relatively small changes, and has specific advantages that it does not have the echoey artefacts that make the “standard” setting useless for percussive instruments, and also the benefit that the length is more or less exact. However, as you have discovered, the “standard” algorithm is better in some cases, which is why it was retained as an option.

An alternative approach, which could be used in conjunction with Change Tempo, is to trim down the gaps between words.
Here’s an example using these settings:
window-Truncate Silence-000.png
and the resulting audio:

Thanks. I found that changing the tempo to 108% produced the best effect on most tracks.

Using the regular algorithm, without high quality, I found that the timing was exact. I sped up video in Camtasia by the same 108%, and then loaded the audio track sped up by Audacity, and they matched exactly. So there may be situations where using the regular algorithm does not produce exact timing, but I did not find any, and the effect would presumably be very small.

Yes. The length error is typically less than 1/10th of a second, which is likely to be insignificant on longer tracks (but unsuitable if you want to tune a single note in a piece of music).